JPH06308997A - Voice synthesizing method - Google Patents

Abstract

PURPOSE:To reduce deterioration quality caused by a change of a pitch frequency at the time of synthesizing a voice by editing a voice waveform. CONSTITUTION:Based on an input phoneme and an input pitch frequency, a voice waveform is selected from a waveform dictionary (S1), and from both a pitch frequency of its selected waveform, and the input pitch frequency, a target formant frequency is determined by referring to a change table (S2). Also, the pitch frequency of the selected waveform is changed in accordance with the input pitch frequency (S3). This format frequency of the voice waveform obtained by changing the pitch frequency is changed to the target formant frequency (S4).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is applied, for example, to the conversion of arbitrary text into speech, selects a speech waveform according to an input phoneme and an input pitch frequency, and determines the pitch frequency of the selected speech waveform as the input pitch frequency. The present invention relates to a voice synthesizing method for controlling a voice in accordance with the input phoneme duration, and setting the length of the selected voice waveform according to the input phoneme duration and sequentially superimposing the waveform as power according to the input magnitude.

[0002]

2. Description of the Related Art As a conventional speech synthesis technique, LPC is used.
A method of synthesizing a voice using a spectrum parameter obtained by analysis and a driving sound source signal such as a pulse has been generally used. This method models a voice and expresses the voice with a finite number of parameters, and it is possible to easily analyze and synthesize the voice by following a certain procedure. There is also a method of selecting an appropriate waveform using a waveform dictionary and performing pitch control using a waveform superposition method to generate a synthesized voice (Japanese Patent Laid-Open No. 01-284898). In this method, since the waveform of natural voice is used as it is, the quality of voice is good,
It is possible to obtain voice equivalent to natural voice.

[0003]

The former method using the spectrum parameter and the driving sound source signal has a problem such as poor voice quality. Also, in the latter edit synthesis, since the correlation between the pitch frequency and the spectral structure is not taken into consideration, the pitch frequency changing process causes quality deterioration in a part of the voice.

An object of the present invention is to provide a method of synthesizing a synthesized voice by performing pitch control using a waveform superposition method, which provides a speech synthesizing method in which quality deterioration due to pitch frequency changing processing is reduced.

[0005]

According to the present invention, a target formant frequency is determined according to an input pitch frequency and a pitch frequency of a selected waveform in a method of generating a synthesized sound by performing pitch control using a waveform superposition method. Then, the formant frequency of the pitch-controlled voice waveform is converted according to the target formant frequency.

[0006]

1 is a flow chart of an embodiment of the method of the present invention. In the present invention, a phoneme (input phoneme) and a pitch frequency (input pitch frequency) are input. For example, when converting text into speech, the text is analyzed and made into a phoneme sequence, and the pitch frequency for each phoneme is further input. And a pitch pattern that considers the continuity of the pitch frequency between phonemes, a phoneme duration for each phoneme, a power (magnitude) for each phoneme, and a power pattern that considers the continuity between phonemes. Is set.

In the main part of the present invention, each input phoneme that is a phoneme sequence and the input pitch frequency for each phoneme are mainly related. That is, first, in the waveform selection step S1, a waveform is selected from the waveform dictionary using an appropriate evaluation function based on the input phoneme and the input pitch frequency. Here, regarding the evaluation function and the waveform selection method, for example, a waveform selection method using the evaluation function (reference “Waveform selection method in waveform editing type regular speech synthesis method”, Sound lecture 1-2-21 (1988)
-10) Hirokawa et al. Can be used. In this method, the optimum waveform from the waveform dictionary, that is, the smallest W, is selected from the waveform dictionary based on the evaluation function W as shown in equation (1) with the target phoneme type, the pitch within the phoneme, the duration, the phoneme match rate, etc. as parameters. Select the waveform to be.

[0008] _{W = Σω i | Pi t -Pi} s | (1) Σ weighting coefficients ωi is the i-th parameter i = 1 to n, Pi _t is a target value of the i-th parameter (input value), Pi _s is the actual value of the i-th parameter (the value of the waveform in the waveform dictionary). Next, in the formant frequency setting step S2, the formant frequency is determined from the input pitch frequency and the pitch frequency of the selected waveform using a table as shown in FIG. For example, if the pitch frequency of the selected waveform is P1 (P _i-1 ≤P1 <P _i ), and it is changed to the input pitch frequency P2 (P _j-1 ≤P2 <P _j ), the _first frequency from FIG. The formant frequency F1 _t is set as F1 _t = F1 ₀ + f1 _ij [Hz] (F1 ₀ is the first formant frequency of the selected waveform) (2). Such a table is prepared, for example, from the first formant to the third formant which are considered to be greatly influenced by the pitch, and the table is referred to for each formant based on the pitch frequency and the input pitch frequency of the selected waveform. Determine the formant frequency.

In the pitch control step S3, the pitch frequency of the selected waveform is changed according to the input pitch frequency. Here, as the pitch frequency changing method, for example, a waveform superimposing method (reference document “Examination of pitch control method in waveform editing type rule synthesis method”, Sound Lecture 1-4-7 (1990-3)
Hirokawa et al.) Can be used. In this method, assuming that the input pitch period is T as shown in FIG. 3, the length is 2T and the front portion and the rear portion gradually decrease with respect to the central portion. Waveform 12 selected in 1 pitch units using window 11
Is cut out, that is, cut out using the window 11 for each pitch period T centered on the large peak of the waveform 12, and these waveforms are superposed at the input pitch period T to obtain the waveform 13 of the input pitch period.

Next, with respect to the pitch control waveform obtained in step S3, in the formant conversion step S4, a waveform in which the formant frequency is changed so as to have the target formant frequency determined in the formant frequency setting step S2 is generated. Here, for example, the formant is changed by using the "formant control method" (Japanese Patent Application No. 4-261825). That is, as shown in FIG. 4, the spectral envelope (spectral density function) P (w) of the selected waveform whose pitch is controlled in the pitch control step S3 is extracted (S ₁ ), and the speech waveform is obtained by the fast Fourier transform (FFT). Is converted into a spectrum X (w) (S ₂ ), and formants are extracted (S ₃ ). Frequency for the purpose of frequency of the formant in which the extracted, i.e. to change the purpose formant frequency determined by the formant frequency setting step S2 (S _4). A spectrum envelope corresponding to the speech waveform in which the formant frequency is changed to the target frequency is calculated for the pitch-controlled selected speech waveform (S ₅ ). The distortion D between the spectral density P ′ (2πΔT · F ′ _i ) at the converted formant frequency F ′ _i and the desired spectral density Pt (2πΔT · F ′ _i ) is obtained (S ₆ ), and the distortion D is sufficiently small. If not (S ₇ ),
Change the bandwidth of the target formant to be changed
Return to ₅ (S ₈ ). When the distortion D is determined to be sufficiently small in step S ₆ , the spectrum X ′ (w) corresponding to the spectrum envelope P ′ (w) finally obtained in step S ₅ from P (w) and X (w). ) Is obtained (S ₉ ), and this X ′ (w) and X (w)
(S ₁₀ ), if the distortion d becomes sufficiently small (S ₁₁ ), X ′ (w) is set to X (w), and X ′
The spectral envelope P ″ (w) of (w) is set as P ′ (w) and the process returns to step S ₉ (S ₁₂ ). If this is repeated and the distortion d becomes sufficiently small, it is finally obtained in step S _9. inverse FFT of X '(w) which are converted into speech waveform (S _13).

With respect to the waveform thus selected, the pitch frequency is changed to the input pitch frequency, the formant frequency of the waveform whose pitch frequency is changed is changed to the target formant frequency, and a voice waveform is obtained. , The phoneme duration is set according to the input text and the power of each phoneme and the power pattern considering the continuity between the phonemes are controlled by controlling the length and power of each phoneme, and the waveforms are sequentially superimposed and synthesized. Get the voice.

The present invention is applicable not only to the case of converting text into speech but also to the case of selecting a speech waveform according to an input phoneme and an input pitch frequency to synthesize speech.

[0013]

As described above, according to the present invention, the pitch frequency of the voice waveform is set to the target frequency in consideration of the correlation between the pitch frequency of the natural voice and the spectral structure, and both pitch frequencies are adjusted. Since the formant frequency is changed by changing the formant frequency, that is, the spectrum is changed, a voice with a changed pitch frequency can be obtained without deteriorating the voice quality, and a synthesized voice with a quality close to natural voice can be obtained.

[Brief description of drawings]

FIG. 1 is a flow chart showing an embodiment of the present invention.

FIG. 2 is a diagram showing a formant frequency change table.

FIG. 3 is a waveform diagram showing an outline of pitch frequency changing processing.

FIG. 4 is a flowchart showing an example of formant conversion processing.

Claims (1)

[Claims] 1. A voice waveform is selected according to an input phoneme and an input pitch frequency, a pitch frequency of the selected voice waveform is controlled according to the input pitch frequency, and the selected voice waveform is controlled according to an input phoneme duration. A target formant frequency is determined according to the input pitch frequency and the pitch frequency of the selected waveform in a speech synthesis method in which waveforms are sequentially superimposed to obtain synthesized speech as power according to the input magnitude. And a formant conversion step of converting the formant frequency of the pitch-controlled voice waveform according to the target formant frequency.